[Mesa-dev] [PATCH 0/4] nir: add support for extended conversions ops needed by OpenCL

Karol Herbst kherbst at redhat.com
Sat Apr 28 11:14:11 UTC 2018


This series adds a bunch of new conversion ops to nir to support all possible
combinations of conversions mandated by the OpenCL specification.

Current drivers shouldn't be affected by any of those changes except renaming
nir_op_f2f16_undef to nir_op_f2f16.

Karol Herbst (4):
  nir: rename f2f16_undef to f2f16
  nir: enable float rounding modes for all float conversions
  nir: add all combinations of conversions with rounding and saturation
  nir: add helper functions for rounding mode

 src/compiler/glsl/glsl_to_nir.cpp |   2 +-
 src/compiler/nir/nir.h            |   2 +-
 src/compiler/nir/nir_opcodes.py   |  30 ++++++-----
 src/compiler/nir/nir_opcodes_c.py |  32 ++++++-----
 src/compiler/nir/nir_opcodes_h.py |  57 ++++++++++++++++++++
 src/compiler/spirv/spirv_to_nir.c |   4 +-
 src/compiler/spirv/vtn_alu.c      | 108 ++++++++++++++++++++++++--------------
 src/compiler/spirv/vtn_glsl450.c  |   2 +-
 src/compiler/spirv/vtn_private.h  |   2 +-
 src/intel/compiler/brw_fs_nir.cpp |   2 +-
 10 files changed, 169 insertions(+), 72 deletions(-)

-- 
2.14.3



More information about the mesa-dev mailing list